AnswersPublished22 sources
Inkling 2 Small: What to Know About Thinking Machines Lab’s New Model
Inkling 2 Small launched on November 5, 2026, as a 594B parameter model with 17B active per token. Artificial Analysis gave it an Intelligence Index score of 31, 18 points above its predecessor; claims of open weight leadership are conditional on the weights being released.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Inkling 2 Small launched on November 5, 2026, as a 594B parameter model with 17B active per token.
- Artificial Analysis gave it an Intelligence Index score of 31, 18 points above its predecessor; claims of open weight leadership are conditional on the weights being released.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What are the key details of Thinking Machines Lab’s November 5 launch of Inkling 2 Small—including the startup’s founding by former OpenAI C. Article summary: Thinking Machines Lab launched Inkling 2 Small on November 5, 2026, as the first model of its second generation, initially offering closed-weight access through Tinker rather than downloadable open weights. [9] The avail. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
Thinking Machines Lab launched Inkling 2 Small on November 5, 2026, as the first model in its second generation. The model has 594 billion total parameters, with 17 billion active per token, and is initially available through the company’s Tinker platform with its weights closed. The company intends to publish the full weights under Apache 2.0, but says that release depends on external safety testing. 3
What is Inkling 2 Small?
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, describes Inkling 2 Small as its most capable model and the first of its second-generation Inkling family. 5
13 It is a mixture-of-experts model that accepts text and image inputs and supports variable reasoning effort. Developers can sample and fine-tune it through Tinker, but cannot download its weights at launch.
2
3
The model’s stated context length is reported as up to one million tokens. That specification should not be confused with the context length available in every Tinker configuration; reporting notes that platform limits may differ. 4
9
Availability and the planned open-weight release
At launch, Inkling 2 Small is hosted on Tinker with closed weights. Thinking Machines says it intends to release the full weights under the Apache 2.0 license after external safety testing and red-teaming. 3
The timing is not settled in the available sources: the launch announcement says “next week,” while the model-family page says “in the coming weeks” and makes the release contingent on additional safety testing. 3
13 Treat either timeframe as a plan, not a confirmed download date.
What the benchmark results say
Artificial Analysis assigns Inkling 2 Small an Intelligence Index score of 31, an 18-point increase over its predecessor. 20 The organization says that, if the weights are released, the model would rank as the highest-scoring open-weight model developed outside China in its current rankings. That is a conditional ranking—not a claim that the model was open-weight at launch or leads every benchmark.
20
A separate report describes its Intelligence Index result as approximately level with Gemini 3.8 Flash at the latter’s highest reasoning setting. 11 That comparison applies to the index result, not necessarily to every task or use case.
Artificial Analysis also reports improved results over the previous Inkling on agent-oriented tests: AutomationBench-AA rose from 5% to 46%, while FrontierCode 2.0 rose from 5% to 25%. 20 The model card lists 13.1% on Terminal-Bench v4.0 and 86.3% on SWE-Bench Pro v2 (Hard); its table marks the latter as the second-highest reported score in that row.
26 These figures come from different evaluations and should not be read as a single, universal measure of coding ability.
Training and practical caveats
Reporting on the launch describes the model’s pretraining as using Nvidia’s NVFP4 four-bit format natively. 17 This is a reported training detail, distinct from the model’s parameter count and evaluation scores.
The release is multimodal in its inputs, but available descriptions say it accepts text and images and produces text—not audio input. 1
6 For developers deciding whether to try it, the immediate practical distinction is access: Tinker supports sampling and fine-tuning, while downloadable weights are still pending safety testing.
2
3
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Inkling 2 Small: What to Know About Thinking Machines Lab’s New Model"?
Inkling 2 Small launched on November 5, 2026, as a 594B parameter model with 17B active per token.
What are the key points to validate first?
Inkling 2 Small launched on November 5, 2026, as a 594B parameter model with 17B active per token. Artificial Analysis gave it an Intelligence Index score of 31, 18 points above its predecessor; claims of open weight leadership are conditional on the weights being released.